This project solves a very broad range of tasks:
At first, we thought all placards will be visible to everyone. But then, it struck us - people may not be comfortable in sharing their name or social media profiles in all surroundings/times. For example, someone may want to share their information only when attending a convention/hackathon/conference.
For solving this, we took inspiration from what WhatsApp did with read receipts. So, we will provide all users with a functionality to toggle their information sharing at any time. And only when a user has toggled it on, they will be able to detect other peoples' information - and when a user has toggled it off, they can only see public placards like that of objects.
We tried out a couple of face detection, recognition, verification and one-shot learning APIs (Azure, AWS Rekognition, Facebook FaceMatch) - however, the challenge was not with this task itself. The actual issue cropped up when we had to integrate these face-to-person mapping data with the AR rendering, as AR requires a 3d model (and not a 2d face image) to render the placard in 3d space.
PS - And due to the 24 hours time constraint, we weren't able to get these face detection results to work with AR. But we could definitely make it work without AR, by just rendering list of people and their contact details in front-end - provided more time.
Discussion